For a project, I've trained multiple networks for multiclass classification all ending with a ReLU activation at the output.
Now the output logits are not probabilities.
Is it valid to get the probability of each class by applying a softmax function at the end, after training?
